Self-Driving Car Engineer Nanodegree¶
Deep Learning¶
Project: Build a Traffic Sign Recognition Classifier¶
In this notebook, a template is provided for you to implement your functionality in stages, which is required to successfully complete this project. If additional code is required that cannot be included in the notebook, be sure that the Python code is successfully imported and included in your submission if necessary.
Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.
In addition to implementing code, there is a writeup to complete. The writeup should be completed in a separate file, which can be either a markdown file or a pdf document. There is a write up template that can be used to guide the writing process. Completing the code template and writeup template will cover all of the rubric points for this project.
The rubric contains "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. The stand out suggestions are optional. If you decide to pursue the "stand out suggestions", you can include the code in this Ipython notebook and also discuss the results in the writeup file.
Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. In addition, Markdown cells can be edited by typically double-clicking the cell to enter edit mode.
Step 0: Load The Data¶
# Load pickled data
import pickle
# TODO: Fill this in based on where you saved the training and testing data
training_file = 'traffic-signs-data/train.p'
validation_file='traffic-signs-data/valid.p'
testing_file = 'traffic-signs-data/test.p'
with open(training_file, mode='rb') as f:
train = pickle.load(f)
with open(validation_file, mode='rb') as f:
valid = pickle.load(f)
with open(testing_file, mode='rb') as f:
test = pickle.load(f)
X_train, y_train = train['features'], train['labels']
X_valid, y_valid = valid['features'], valid['labels']
X_test, y_test = test['features'], test['labels']
Step 1: Dataset Summary & Exploration¶
The pickled data is a dictionary with 4 key/value pairs:
'features'is a 4D array containing raw pixel data of the traffic sign images, (num examples, width, height, channels).'labels'is a 1D array containing the label/class id of the traffic sign. The filesignnames.csvcontains id -> name mappings for each id.'sizes'is a list containing tuples, (width, height) representing the original width and height the image.'coords'is a list containing tuples, (x1, y1, x2, y2) representing coordinates of a bounding box around the sign in the image. THESE COORDINATES ASSUME THE ORIGINAL IMAGE. THE PICKLED DATA CONTAINS RESIZED VERSIONS (32 by 32) OF THESE IMAGES
Complete the basic data summary below. Use python, numpy and/or pandas methods to calculate the data summary rather than hard coding the results. For example, the pandas shape method might be useful for calculating some of the summary results.
Provide a Basic Summary of the Data Set Using Python, Numpy and/or Pandas¶
### Replace each question mark with the appropriate value.
### Use python, pandas or numpy methods rather than hard coding the results
import numpy as np
# TODO: Number of training examples
n_train = X_train.shape[0]
# TODO: Number of validation examples
n_validation = X_valid.shape[0]
# TODO: Number of testing examples.
n_test = X_test.shape[0]
# TODO: What's the shape of an traffic sign image?
image_shape = (X_train.shape[1],X_train.shape[2])
# TODO: How many unique classes/labels there are in the dataset.
n_classes = np.min(y_train)+np.max(y_train)+1
print("Number of training examples =", n_train)
print("Number of testing examples =", n_test)
print("Image data shape =", image_shape)
print("Number of classes =", n_classes)
Include an exploratory visualization of the dataset¶
Visualize the German Traffic Signs Dataset using the pickled file(s). This is open ended, suggestions include: plotting traffic sign images, plotting the count of each sign, etc.
The Matplotlib examples and gallery pages are a great resource for doing visualizations in Python.
NOTE: It's recommended you start with something simple first. If you wish to do more, come back to it after you've completed the rest of the sections. It can be interesting to look at the distribution of classes in the training, validation and test set. Is the distribution the same? Are there more examples of some classes than others?
### Data exploration visualization code goes here.
### Feel free to use as many code cells as needed.
import matplotlib.pyplot as plt
import matplotlib.cm as cm
import random
# Visualizations will be shown in the notebook.
%matplotlib inline
def visualize_dataset(features,labels,signnames,label_no):
label_name = signnames.index[label_no]
print('class %s: %s' %(label_no, label_name))
n_img_to_visual=20
n_row_plot = 2
n_col_plot = int(n_img_to_visual/n_row_plot)
index = np.where(labels==label_no)
label_indexes = index[0][0:n_img_to_visual]
#fig = plt.figure()
fig = plt.figure(figsize = (n_col_plot, n_row_plot))
plot_count = 0
fig.subplots_adjust(hspace = 0, wspace = 0)
for i in range(n_col_plot*n_row_plot):
ax = fig.add_subplot(n_row_plot, n_col_plot, plot_count+1)
ax.axis('off')
plot_count += 1
idx = label_indexes[i]
img = np.squeeze(features[idx,:,:,:])
if img.ndim==2:
plt.imshow(img, cmap=cm.gray)
else:
plt.imshow(img)
plt.show()
#read the traffic sign name mapping
import pandas as pd
signnames = pd.read_csv('signnames.csv', index_col='SignName')
#show the images in all classes
for label_no in range(n_classes):
visualize_dataset(X_train,y_train,signnames,label_no)
#show the distribution of training set
fig= plt.plot()
plt.hist(y_train, normed=True, bins=n_classes)
plt.xlabel('class no.')
plt.ylabel('Probability')
plt.title('training set distribution')
plt.show()
#show the distribution of validation set
fig= plt.plot()
plt.hist(y_valid, normed=True, bins=n_classes)
plt.xlabel('class no.')
plt.ylabel('Probability')
plt.title('validation set distribution')
plt.show()
#show the distribution of testing set
fig= plt.plot()
plt.hist(y_test, normed=True, bins=n_classes)
plt.xlabel('class no.')
plt.ylabel('Probability')
plt.title('testing set distribution')
plt.show()
Step 2: Design and Test a Model Architecture¶
Design and implement a deep learning model that learns to recognize traffic signs. Train and test your model on the German Traffic Sign Dataset.
The LeNet-5 implementation shown in the classroom at the end of the CNN lesson is a solid starting point. You'll have to change the number of classes and possibly the preprocessing, but aside from that it's plug and play!
With the LeNet-5 solution from the lecture, you should expect a validation set accuracy of about 0.89. To meet specifications, the validation set accuracy will need to be at least 0.93. It is possible to get an even higher accuracy, but 0.93 is the minimum for a successful project submission.
There are various aspects to consider when thinking about this problem:
- Neural network architecture (is the network over or underfitting?)
- Play around preprocessing techniques (normalization, rgb to grayscale, etc)
- Number of examples per label (some have more than others).
- Generate fake data.
Here is an example of a published baseline model on this problem. It's not required to be familiar with the approach used in the paper but, it's good practice to try to read papers like these.
Pre-process the Data Set (normalization, grayscale, etc.)¶
Minimally, the image data should be normalized so that the data has mean zero and equal variance. For image data, (pixel - 128)/ 128 is a quick way to approximately normalize the data and can be used in this project.
Other pre-processing steps are optional. You can try different techniques to see if it improves performance.
Use the code cell (or multiple code cells, if necessary) to implement the first step of your project.
### Preprocess the data here. It is required to normalize the data. Other preprocessing steps could include
### converting to grayscale, etc.
### Feel free to use as many code cells as needed.
import matplotlib.pyplot as plt
import numpy as np
import skimage.transform as sk_tf
from skimage.util import random_noise
from skimage.exposure import equalize_hist
class Data_Augment(object):
def __init__(self):
pass
#Rotation angle in degrees in counter-clockwise direction.
def rotate(self,images, rot_deg):
augment_images = np.copy(images)*0.0
for i in range(len(images)):
augment_images[i,:,:,:] = sk_tf.rotate(images[i,:,:,:],rot_deg)
return augment_images
#histogram equalization
def hist_equal(self,images):
augment_images = np.copy(images)*0.0
for i in range(len(images)):
augment_images[i,:,:,:] = equalize_hist(images[i,:,:,:])
return augment_images
#add gaussian noise
def noise(self,images):
augment_images = np.copy(images)*0.0
for i in range(len(images)):
augment_images[i,:,:,:] = random_noise(images[i,:,:,:],mode='gaussian')
return augment_images
#affine images by image scale, translation, rotation
def affine(self,images,scale=(1.0, 1.0), rotation=0.0,translation=(0, 0)):
augment_images = np.copy(images)*0.0
shift_form = sk_tf.AffineTransform(scale=scale, rotation=rotation,translation=translation)
for i in range(len(images)):
augment_images[i,:,:,:]=sk_tf.warp(images[i,:,:,:],shift_form)
return augment_images
Show Augmentation images¶
augmented images inclued noisy, histogram equalization, transition, zoom up, and rotation images
import warnings
warnings.filterwarnings('ignore')
#create augment object
Augment = Data_Augment()
#choose a item to test
test_item = 500
test_image = np.expand_dims(X_train[test_item,:,:,:],axis=0)
#Start to augmentate!
#add noise
noise_image = np.squeeze(Augment.noise(test_image))
#histgorm equalization
hist_equal_image = np.squeeze(Augment.hist_equal(test_image))
#translation shift-up
affine_image1 = np.squeeze(Augment.affine(test_image,scale=(1.0, 1.0),rotation=0.0,translation=(0, 1)))
#translation shoft-down
affine_image2 = np.squeeze(Augment.affine(test_image,scale=(1.0, 1.0),rotation=0.0,translation=(0, -1)))
#translation shift-left
affine_image3 = np.squeeze(Augment.affine(test_image,scale=(1.0, 1.0),rotation=0.0,translation=(1, 0)))
#translation shift-right
affine_image4 = np.squeeze(Augment.affine(test_image,scale=(1.0, 1.0),rotation=0.0,translation=(-1,0)))
#scale x up
affine_image5 = np.squeeze(Augment.affine(test_image,scale=(0.9, 1.0),rotation=0.0,translation=(0, 0)))
#scale y up
affine_image6 = np.squeeze(Augment.affine(test_image,scale=(1.0, 0.9),rotation=0.0,translation=(0, 0)))
#scale x and y up
affine_image7 = np.squeeze(Augment.affine(test_image,scale=(0.9, 0.9),rotation=0.0,translation=(0, 0)))
#scale x and y up
affine_image8 = np.squeeze(Augment.affine(test_image,scale=(0.85, 0.85),rotation=0.0,translation=(0, 0)))
#rotate
rotate_image1 = np.squeeze(Augment.rotate(test_image,4))
rotate_image2 = np.squeeze(Augment.rotate(test_image,-4))
rotate_image3 = np.squeeze(Augment.rotate(test_image,2))
rotate_image4 = np.squeeze(Augment.rotate(test_image,-2))
#plot augmentation images
n_col_plot = 5
n_row_plot = 3
fig = plt.figure(figsize = (2*n_col_plot, 3*n_row_plot))
plot_count = 0
fig.subplots_adjust(hspace = 0, wspace = 0)
ax = fig.add_subplot(n_row_plot, n_col_plot, 1)
ax.axis('off')
plt.title('origin')
plt.imshow(np.squeeze(test_image))
ax = fig.add_subplot(n_row_plot, n_col_plot, 2)
ax.axis('off')
plt.title('noise')
plt.imshow(noise_image)
ax = fig.add_subplot(n_row_plot, n_col_plot, 3)
ax.axis('off')
plt.title('hist equal')
plt.imshow(hist_equal_image)
ax = fig.add_subplot(n_row_plot, n_col_plot, 4)
ax.axis('off')
plt.title('shift-up:')
plt.imshow(affine_image1)
ax = fig.add_subplot(n_row_plot, n_col_plot, 5)
ax.axis('off')
plt.title('shift-down')
plt.imshow(affine_image2)
ax = fig.add_subplot(n_row_plot, n_col_plot, 6)
ax.axis('off')
plt.title('shift-left')
plt.imshow(affine_image3)
ax = fig.add_subplot(n_row_plot, n_col_plot, 7)
ax.axis('off')
plt.title('shift-right')
plt.imshow(affine_image4)
ax = fig.add_subplot(n_row_plot, n_col_plot, 8)
ax.axis('off')
plt.title('scale x')
plt.imshow(affine_image5)
ax = fig.add_subplot(n_row_plot, n_col_plot, 9)
ax.axis('off')
plt.title('scale y')
plt.imshow(affine_image6)
ax = fig.add_subplot(n_row_plot, n_col_plot, 10)
ax.axis('off')
plt.title('scale x/y')
plt.imshow(affine_image7)
ax = fig.add_subplot(n_row_plot, n_col_plot, 11)
ax.axis('off')
plt.title('scale x/y')
plt.imshow(affine_image8)
ax = fig.add_subplot(n_row_plot, n_col_plot, 12)
ax.axis('off')
plt.title('rotate 4 degree')
plt.imshow(rotate_image1)
ax = fig.add_subplot(n_row_plot, n_col_plot, 13)
ax.axis('off')
plt.title('rotate -4 degree')
plt.imshow(rotate_image2)
ax = fig.add_subplot(n_row_plot, n_col_plot, 14)
ax.axis('off')
plt.title('rotate 2 degree')
plt.imshow(rotate_image3)
ax = fig.add_subplot(n_row_plot, n_col_plot, 15)
ax.axis('off')
plt.title('rotate -2 degree')
plt.imshow(rotate_image4)
plt.show()
Augment train data set¶
#add noise
noise_train = Augment.noise(X_train)
#histogram equalization
hist_equal_train = Augment.hist_equal(X_train)
#rotate
rotate_train1 = Augment.rotate(X_train,2)
rotate_train2 = Augment.rotate(X_train,-2)
#rotate_train3 = Augment.rotate(X_train,4)
#rotate_train4 = Augment.rotate(X_train,-4)
#affine
affine_train1 = Augment.affine(X_train,scale=(0.9,0.9),rotation=0.0,translation=(0,0))
affine_train2 = Augment.affine(X_train,scale=(0.9,1.0),rotation=0.0,translation=(0,0))
affine_train3 = Augment.affine(X_train,scale=(1.0,0.9),rotation=0.0,translation=(0,0))
#affine_train3 = Augment.affine(X_train,scale=(0.85,0.85),rotation=0.0,translation=(0,0))
#affine_train5 = Augment.affine(X_train,scale=(1.0,1.0),rotation=0.0,translation=(1,0))
#affine_train6 = Augment.affine(X_train,scale=(1.0,1.0),rotation=0.0,translation=(-1,0))
#affine_train7 = Augment.affine(X_train,scale=(1.0,1.0),rotation=0.0,translation=(0,1))
#affine_train8 = Augment.affine(X_train,scale=(1.0,1.0),rotation=0.0,translation=(0,-1))
#class_to_visual = 0
#visualize_dataset(noise_train,y_train,signnames,class_to_visual)
#combine augmentation data into training set
Augment_X_train = np.concatenate((X_train,noise_train,hist_equal_train,rotate_train1,rotate_train2,affine_train1,affine_train2,affine_train3),axis=0)
Augment_y_train = np.concatenate((y_train,y_train,y_train,y_train,y_train,y_train,y_train,y_train),axis=0)
print(Augment_X_train.shape)
print(Augment_y_train.shape)
def normalize_images(images,mode='zero_mean'):
norm_images = np.copy(images)*0.0
if mode=='zero_median':
for i in range(len(images)):
norm_images[i,:,:,:] = (images[i,:,:,:]-128.0)/128.0
elif mode=='zero_mean':
for i in range(len(images)):
i_std = np.std(images[i,:,:,:])
i_mean = np.mean(images[i,:,:,:])
norm_images[i,:,:,:] = (images[i,:,:,:]-i_mean)/(i_std+0.000001)
elif mode=='min_max':
for i in range(len(images)):
i_min = np.min(images[i,:,:,:])
i_max = np.max(images[i,:,:,:])
norm_images[i,:,:,:] = (images[i,:,:,:]-i_min)/(i_max-i_min+0.000001)
else:
norm_images = np.copy(images)
return norm_images
def show_data_analysis(features):
print('--------------------------------------------------------')
print(' mean of the dataset = %s' %(np.mean(features)))
print(' std of the dataset = %s' %(np.std(features)))
print(' range of the dataset(min,max) = (%s , %s)' %(np.min(features),np.max(features)))
print('--------------------------------------------------------')
print('train data info before normalization')
show_data_analysis(Augment_X_train)
norm_X_train = normalize_images(Augment_X_train)
print('train data info after normalization')
show_data_analysis(norm_X_train)
norm_y_train = np.copy(Augment_y_train)
#clear memory
del Augment_X_train
del Augment_y_train
del noise_train
del hist_equal_train
del rotate_train1
del rotate_train2
del affine_train1
del affine_train2
del affine_train3
Model Architecture¶
define the basic convolution network layer
import tensorflow as tf
#define the basic convolution and fc layer in cnn
def cnn_conv_layer(name,input,filters,kernel_size=[5,5],w_mu=0.0,w_sigma=0.1,padding='VALID',activation=None):
with tf.name_scope(name):
conv_W = tf.Variable(tf.truncated_normal(shape=(kernel_size[0], kernel_size[1], input.get_shape().as_list()[3], filters), mean = w_mu, stddev = w_sigma),name='conv_weight')
conv_b = tf.Variable(tf.zeros([filters]),name='conv_bias')
conv = tf.nn.conv2d(input, conv_W, strides=[1, 1, 1, 1], padding=padding) + conv_b
if activation is not None:
conv = activation(conv)
return conv
def cnn_fc_layer(name,input,filters,w_mu=0.0,w_sigma=0.1,activation=None):
fc_W = tf.Variable(tf.truncated_normal(shape=(input.get_shape().as_list()[1] , filters), mean = w_mu, stddev = w_sigma),name='fc_weight')
fc_b = tf.Variable(tf.zeros([filters]),name='fc_bias')
fc = tf.nn.xw_plus_b(input, fc_W, fc_b)
if activation is not None:
fc = activation(fc)
return fc
def max_pool(x,filter_size,stride_size, padding='VALID'):
x = tf.nn.max_pool(x, ksize=[1, filter_size, filter_size, 1], strides=[1, stride_size, stride_size, 1], padding=padding)
return x
Model for traffic classifier¶
The model name is from my cat shoo-shoo ^^
from tensorflow.contrib.layers import flatten
def ShooNet(x,keep_prob,no_classes):
#Hyperparameters
mu = 0.0
sigma = 0.1
with tf.name_scope('ShooNet'):
# Convolutional Layer Input = 32x32x3. Output = 28x28x48.
conv1_act = cnn_conv_layer('conv1',x,48,kernel_size=[5,5],w_mu=mu,w_sigma=sigma,padding='VALID',activation=tf.nn.relu)
# Max Pooling. Input = 28x28x48. Output = 14x14x48
conv1_pl = max_pool(conv1_act,2,2,padding='VALID')
# Dropout
conv1_dp = tf.nn.dropout(conv1_pl,keep_prob)
# Convolutional Layer. Output = 10x10x96.
conv2_act = cnn_conv_layer('conv2',conv1_dp,96,kernel_size=[5,5],w_mu=mu,w_sigma=sigma,padding='VALID',activation=tf.nn.relu)
# Max Pooling. Input = 10x10x96. Output = 5x5x96
conv2_pl = max_pool(conv2_act,2,2,padding='VALID')
# Dropout
conv2_dp = tf.nn.dropout(conv2_pl,keep_prob)
# Convolutional Layer. Input = 5x5x96. Output = 3x3x172
conv3_act = cnn_conv_layer('conv3', conv2_dp,172,kernel_size=[3,3],w_mu=mu,w_sigma=sigma,padding='VALID',activation=tf.nn.relu)
# Max Pooling. Input = 3x3x172. Output = 2x2x172.
conv3_pl = max_pool(conv3_act,2,1,padding='VALID')
# Dropout
conv3_dp = tf.nn.dropout(conv3_pl,keep_prob)
# Flatten. Input = 2x2x172. Output = 688.
fc1 = flatten(conv3_dp)
# Fully Connected. Input = 688. Output = 344.
fc2 = cnn_fc_layer('fc2',fc1,344,w_mu=0.0,w_sigma=0.1,activation=tf.nn.relu)
fc2_dp = tf.nn.dropout(fc2,keep_prob)
# Fully Connected. Input = 344. Output = 172.
fc3 = cnn_fc_layer('fc3',fc2_dp,172,w_mu=0.0,w_sigma=0.1,activation=tf.nn.relu)
fc3_dp = tf.nn.dropout(fc3,keep_prob)
# Fully Connected. Input = 172. Output = 86.
fc4 = cnn_fc_layer('fc4',fc3_dp,86,w_mu=0.0,w_sigma=0.1,activation=tf.nn.relu)
fc4_dp = tf.nn.dropout(fc4,keep_prob)
# Fully Connected. Input = 86. Output = 43.
logits = cnn_fc_layer('logits',fc4_dp,no_classes,w_mu=0.0,w_sigma=0.1,activation=None)
return logits,conv1_act,conv2_act,conv3_act
Loss, accuracy and training operation¶
#placehoder for train set
x = tf.placeholder(tf.float32, (None, 32, 32, 3))
y = tf.placeholder(tf.int32, (None))
#dropout (keep probability)
keep_prob = tf.placeholder(tf.float32)
#encode labels into one-hot
y_one_hot = tf.one_hot(y, n_classes)
#prediction value from the model
logits,conv1,conv2,conv3= ShooNet(x,keep_prob,n_classes)
#define the loss function
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=logits,labels= y_one_hot)
loss_operation = tf.reduce_mean(cross_entropy)
optimizer = tf.train.AdamOptimizer(learning_rate = 0.001)
#define the training operation
training_operation = optimizer.minimize(loss_operation)
#accuracy measurement
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(y_one_hot, 1))
accuracy_operation = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
def evaluate_acc(X_data, y_data):
BATCH_SIZE = 128
num_examples = len(X_data)
total_accuracy = 0
sess = tf.get_default_session()
for offset in range(0, num_examples, BATCH_SIZE):
batch_x, batch_y = X_data[offset:offset+BATCH_SIZE], y_data[offset:offset+BATCH_SIZE]
accuracy = sess.run(accuracy_operation, feed_dict={x: batch_x, y: batch_y,keep_prob:1.0})
total_accuracy += (accuracy * len(batch_x))
return total_accuracy / num_examples
def evaluate_loss(X_data, y_data):
BATCH_SIZE = 128
num_examples = len(X_data)
total_loss = 0
sess = tf.get_default_session()
total_num_of_batch = 0
for offset in range(0, num_examples, BATCH_SIZE):
batch_x, batch_y = X_data[offset:offset+BATCH_SIZE], y_data[offset:offset+BATCH_SIZE]
loss = sess.run(loss_operation, feed_dict={x: batch_x, y: batch_y,keep_prob:1.0})
total_num_of_batch+=1
total_loss +=loss
#total_loss += (loss * len(batch_x))
return total_loss / total_num_of_batch
Train, Validate and Test the Model¶
A validation set can be used to assess how well the model is performing. A low accuracy on the training and validation sets imply underfitting. A high accuracy on the training set but low accuracy on the validation set implies overfitting.
def plot_hist_loss(train_loss_list,valid_loss_list,iter_list):
fig = plt.plot()
plt.plot(iter_list,np.log(train_loss_list),label='train_loss')
plt.plot(iter_list,np.log(valid_loss_list),label='valid_loss')
plt.legend(loc='upper right')
plt.title('loss')
plt.xlabel('iter')
plt.ylabel('log(loss)')
plt.show()
def plot_hist_acc(train_acc_list,valid_acc_list,iter_list):
fig = plt.plot()
plt.plot(iter_list,train_acc_list,label='train_acc')
plt.plot(iter_list,valid_acc_list,label='valid_acc')
plt.legend(loc='lower right')
plt.title('accuracy')
plt.xlabel('iter')
plt.ylabel('accuracy')
plt.show()
### Train your model here.
### Calculate and report the accuracy on the training and validation set.
### Once a final model architecture is selected,
### the accuracy on the test set should be calculated and reported as well.
### Feel free to use as many code cells as needed.
from sklearn.utils import shuffle
#training parameters
EPOCHS =80
BATCH_SIZE = 128
verfiy_step = 100
disp_step = 1000
save_model_step = 500
drop_out = 0.8
#model directory
model_dir = 'model/'
model_path = model_dir + 'shoo.ckpt'
iter_list = []
train_acc_list = []
valid_acc_list = []
saver = tf.train.Saver()
with tf.Session() as sess:
merged = tf.summary.merge_all()
writer = tf.summary.FileWriter('/tmp/tensorflowlogs',sess.graph)
sess.run(tf.global_variables_initializer())
#restore model from last step
ckpt = tf.train.get_checkpoint_state(model_dir)
if ckpt and ckpt.model_checkpoint_path:
print('restore model %s' %ckpt.model_checkpoint_path)
saver.restore(sess, ckpt.model_checkpoint_path)
num_examples = len(X_train)
step = 1
print("Training...")
for i in range(EPOCHS):
print("EPOCH {} ... ".format(i+1))
norm_X_train, norm_y_train = shuffle(norm_X_train, norm_y_train)
for offset in range(0, num_examples, BATCH_SIZE):
end = offset + BATCH_SIZE
batch_x, batch_y = norm_X_train[offset:end], norm_y_train[offset:end]
sess.run(training_operation, feed_dict={x: batch_x, y: batch_y,keep_prob:drop_out})
if step % verfiy_step ==0:
#evaluate accuracy
valid_acc = evaluate_acc(X_valid, y_valid)
train_acc = evaluate_acc(norm_X_train,norm_y_train)
#add accuracy info intothe array
train_acc_list.append(train_acc)
valid_acc_list.append(valid_acc)
iter_list.append(step)
#display accuracy
print("Validation acc = {:.3f}".format(valid_acc), end=" ")
print("Training acc = {:.3f}".format(train_acc))
if step % disp_step == 0:
#plot the history accuracy
plot_hist_acc(train_acc_list,valid_acc_list,iter_list)
if step % save_model_step == 0:
saver.save(sess, model_path, global_step=step)
step +=1
saver.save(sess, model_path)
print("Model saved")
Accuracy in test set¶
### Load the images and plot them here.
### Feel free to use as many code cells as needed.
model_path = 'model/shoo.ckpt'
#normalize testing set
norm_X_test = normalize_images(X_test)
saver = tf.train.Saver()
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
saver.restore(sess,model_path)
test_acc = evaluate_acc(norm_X_test,y_test)
print('Test set accuracy = {:.3f}'.format(test_acc))
Step 3: Test a Model on New Images¶
To give yourself more insight into how your model is working, download at least five pictures of German traffic signs from the web and use your model to predict the traffic sign type.
You may find signnames.csv useful as it contains mappings from the class id (integer) to the actual sign name.
Load and Output the Images¶
import glob
import matplotlib.pyplot as plt
from scipy.misc import imread
from scipy.misc import imresize
import pandas as pd
import cv2
import numpy as np
signnames = pd.read_csv('signnames.csv', index_col='SignName')
image_filename_list = glob.glob('GTSRB/Images/*.ppm')
pd = pd.read_csv('GTSRB/GTSRB_Online-Test-GT/GT-online_test.csv',sep=';')
[w, h, x1, x2, y1, y2, id] = pd['Width'], pd['Height'], pd['Roi.X1'], pd['Roi.X2'], pd['Roi.Y1'], pd['Roi.Y2'],pd['ClassId']
GTSRB_X = []
GTSRB_y = []
num_of_test = 16
for i in range(num_of_test):
img = imread(image_filename_list[i])
img_crop = img[y1[i]:y2[i]+1,x1[i]:x2[i]+1,:]
img_crop_resized= cv2.resize(img_crop, (32,32),interpolation=cv2.INTER_CUBIC)
GTSRB_X.append(img_crop_resized)
GTSRB_y.append(id[i])
GTSRB_X = np.array(GTSRB_X)
GTSRB_y = np.array(GTSRB_y)
#visualize test image
n_row_plot = 4
n_col_plot = int(num_of_test/n_row_plot)
fig = plt.figure(figsize = (5*n_col_plot, 5*n_row_plot))
fig.subplots_adjust(hspace = 0.5, wspace = 0.5)
plot_count = 1
for i in range(num_of_test):
ax = fig.add_subplot(n_row_plot, n_col_plot, plot_count)
ax.axis('off')
label_id = int(GTSRB_y[i])
item_name = signnames.index[label_id]
plt.title('Class Id:%s \n %s' %(label_id,item_name))
plt.imshow(GTSRB_X[i])
plot_count+=1
#normalize test image
norm_GTSRB_X = normalize_images(GTSRB_X)
Predict the Sign Type for Each Image¶
### Run the predictions here and use the model to output the prediction for each image.
### Make sure to pre-process the images with the same pre-processing pipeline used earlier.
### Feel free to use as many code cells as needed.
def inference_Traffic_Classifier(X_data, sess):
traffic_sign = sess.run(tf.argmax(logits, 1), feed_dict={x: X_data, keep_prob: 1.0})
return traffic_sign
def inference_Traffic_Classifier_Top5(X_data,sess):
top5 = sess.run(tf.nn.top_k(tf.nn.softmax(logits), k=5), feed_dict={x: X_data, keep_prob: 1.0})
return top5
def inference_Traffic_Classifier_prob(X_data,sess):
prob = sess.run(tf.nn.softmax(logits), feed_dict={x: X_data, keep_prob: 1.0})
return prob
Predict traffic sign in German Traffic Sign Dataset¶
#normalize testing set
norm_GTSRB_X = normalize_images(GTSRB_X)
model_path = 'model/shoo.ckpt'
saver = tf.train.Saver()
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
saver.restore(sess,model_path)
pred_sign = inference_Traffic_Classifier(norm_GTSRB_X, sess)
#visualize prediction answer
n_row_plot = 4
n_col_plot = int(num_of_test/n_row_plot)
fig = plt.figure(figsize = (5*n_col_plot, 5*n_row_plot))
fig.subplots_adjust(hspace = 0.5, wspace = 0.5)
plot_count = 1
for i in range(num_of_test):
ax = fig.add_subplot(n_row_plot, n_col_plot, plot_count)
ax.axis('off')
item_name = signnames.index[label_id]
plt.title('GT: %s\n Pred: %s' %(signnames.index[GTSRB_y[i]],signnames.index[pred_sign[i]]))
plt.imshow(GTSRB_X[i])
plot_count+=1
Analyze Performance¶
### Calculate the accuracy for these 5 new images.
### For example, if the model predicted 1 out of 5 signs correctly, it's 20% accurate on these new images.
GT_sign = GTSRB_y[0:num_of_test]
print('class id of ground truth: %s' %GT_sign)
print('class id of predictor: %s' %pred_sign)
diff_sign = GT_sign - pred_sign
num_of_hit_item = np.array((np.where(diff_sign==0))).shape[1]
accuracy = num_of_hit_item / num_of_test *100
print('Accuracy of test images is %s %%' %accuracy)
Output Top 5 Softmax Probabilities For Each Image Found on the Web¶
For each of the new images, print out the model's softmax probabilities to show the certainty of the model's predictions (limit the output to the top 5 probabilities for each image). tf.nn.top_k could prove helpful here.
The example below demonstrates how tf.nn.top_k can be used to find the top k predictions for each image.
tf.nn.top_k will return the values and indices (class ids) of the top k predictions. So if k=3, for each sign, it'll return the 3 largest probabilities (out of a possible 43) and the correspoding class ids.
Take this numpy array as an example. The values in the array represent predictions. The array contains softmax probabilities for five candidate images with six possible classes. tf.nn.top_k is used to choose the three classes with the highest probability:
# (5, 6) array
a = np.array([[ 0.24879643, 0.07032244, 0.12641572, 0.34763842, 0.07893497,
0.12789202],
[ 0.28086119, 0.27569815, 0.08594638, 0.0178669 , 0.18063401,
0.15899337],
[ 0.26076848, 0.23664738, 0.08020603, 0.07001922, 0.1134371 ,
0.23892179],
[ 0.11943333, 0.29198961, 0.02605103, 0.26234032, 0.1351348 ,
0.16505091],
[ 0.09561176, 0.34396535, 0.0643941 , 0.16240774, 0.24206137,
0.09155967]])Running it through sess.run(tf.nn.top_k(tf.constant(a), k=3)) produces:
TopKV2(values=array([[ 0.34763842, 0.24879643, 0.12789202],
[ 0.28086119, 0.27569815, 0.18063401],
[ 0.26076848, 0.23892179, 0.23664738],
[ 0.29198961, 0.26234032, 0.16505091],
[ 0.34396535, 0.24206137, 0.16240774]]), indices=array([[3, 0, 5],
[0, 1, 4],
[0, 5, 1],
[1, 3, 5],
[1, 4, 3]], dtype=int32))Looking just at the first row we get [ 0.34763842, 0.24879643, 0.12789202], you can confirm these are the 3 largest probabilities in a. You'll also notice [3, 0, 5] are the corresponding indices.
### Print out the top five softmax probabilities for the predictions on the German traffic sign images found on the web.
### Feel free to use as many code cells as needed.
from pylab import *
#get the top5 prediction
model_path = 'model/shoo.ckpt'
saver = tf.train.Saver()
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
saver.restore(sess,model_path)
pred_Top5 = inference_Traffic_Classifier_Top5(norm_GTSRB_X,sess)
#plot top 5 prediction results
n_row_plot =8
n_col_plot = int(num_of_test/n_row_plot)
fig = plt.figure(figsize = (4*n_col_plot, 4*n_row_plot))
fig.subplots_adjust(hspace = 1.2, wspace = 0.8)
plot_count = 1
for i in range(num_of_test):
signname_idx = []
indices = pred_Top5.indices[i]
values = pred_Top5.values[i]
for s in range(len(indices)):
signname_idx.append(signnames.index[indices[s]])
ax = fig.add_subplot(n_row_plot, n_col_plot, plot_count)
pos = arange(len(signname_idx))+.5 # the bar centers on the y axis
barh(pos,values, align='center')
yticks(pos, signname_idx)
xlabel('Probability')
title('Top 5 : \n %s' %(signnames.index[GTSRB_y[i]]))
grid(True)
plot_count+=1
plt.show()
Note: Once you have completed all of the code implementations and successfully answered each question above, you may finalize your work by exporting the iPython Notebook as an HTML document. You can do this by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.
Step 4 (Optional): Visualize the Neural Network's State with Test Images¶
This Section is not required to complete but acts as an additional excersise for understaning the output of a neural network's weights. While neural networks can be a great learning device they are often referred to as a black box. We can understand what the weights of a neural network look like better by plotting their feature maps. After successfully training your neural network you can see what it's feature maps look like by plotting the output of the network's weight layers in response to a test stimuli image. From these plotted feature maps, it's possible to see what characteristics of an image the network finds interesting. For a sign, maybe the inner network feature maps react with high activation to the sign's boundary outline or to the contrast in the sign's painted symbol.
Provided for you below is the function code that allows you to get the visualization output of any tensorflow weight layer you want. The inputs to the function should be a stimuli image, one used during training or a new one you provided, and then the tensorflow variable name that represents the layer's state during the training process, for instance if you wanted to see what the LeNet lab's feature maps looked like for it's second convolutional layer you could enter conv2 as the tf_activation variable.
For an example of what feature map outputs look like, check out NVIDIA's results in their paper End-to-End Deep Learning for Self-Driving Cars in the section Visualization of internal CNN State. NVIDIA was able to show that their network's inner weights had high activations to road boundary lines by comparing feature maps from an image with a clear path to one without. Try experimenting with a similar test to show that your trained network's weights are looking for interesting features, whether it's looking at differences in feature maps from images with or without a sign, or even what feature maps look like in a trained network vs a completely untrained one on the same sign image.
Your output should look something like this (above)
### Visualize your network's feature maps here.
### Feel free to use as many code cells as needed.
# image_input: the test image being fed into the network to produce the feature maps
# tf_activation: should be a tf variable name used during your training procedure that represents the calculated state of a specific weight layer
# activation_min/max: can be used to view the activation contrast in more detail, by default matplot sets min and max to the actual min and max values of the output
# plt_num: used to plot out multiple different weight feature map sets on the same block, just extend the plt number for each new feature map entry
def outputFeatureMap(image_input, tf_activation, activation_min=-1, activation_max=-1 ,plt_num=1):
# Here make sure to preprocess your image_input in a way your network expects
# with size, normalization, ect if needed
# image_input =
# Note: x should be the same name as your network's tensorflow data placeholder variable
# If you get an error tf_activation is not defined it may be having trouble accessing the variable from inside a function
activation = tf_activation.eval(session=sess,feed_dict={x : image_input,keep_prob:1.0})
featuremaps = activation.shape[3]
plt.figure(plt_num, figsize=(15,15))
for featuremap in range(featuremaps):
plt.subplot(6,16, featuremap+1) # sets the number of feature maps to show on each row and column
plt.title('FeatureMap ' + str(featuremap)) # displays the feature map number
if activation_min != -1 & activation_max != -1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmin =activation_min, vmax=activation_max, cmap="gray")
elif activation_max != -1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmax=activation_max, cmap="gray")
elif activation_min !=-1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmin=activation_min, cmap="gray")
else:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", cmap="gray")
test_item = 11
img = GTSRB_X[test_item,:,:,:]
image_input = np.expand_dims(norm_GTSRB_X[test_item,:,:,:],axis=0)
figure = plt.plot()
plt.imshow(img)
plt.show()
print('visualize the conv layer1')
model_path = 'model/shoo.ckpt'
saver = tf.train.Saver()
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
saver.restore(sess,model_path)
outputFeatureMap(image_input,conv1, activation_min=-1, activation_max=-1 ,plt_num=1)
#show the tensor name in the model
model_path = 'model/shoo.ckpt'
saver = tf.train.Saver()
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
saver.restore(sess,model_path)
graph = tf.get_default_graph()
for v in tf.get_collection(tf.GraphKeys. GLOBAL_VARIABLES):
layer = graph.get_tensor_by_name(v.name)
print('layer name:%s, size = %s' %(v.name,layer.shape))